Before it was said that the United States invented, China copied and Europe regulated. Now what is being said is that China has the data; United States, pasta; and that Europe has a plan . The plan of the European Union (EU), as manifested in its proposal to regulate artificial intelligence and facial recognition systems, is to try to reconcile two aspects that, in principle, seem contradictory: on the one hand, to protect the rights of European citizens and, on the other, accelerate the development of artificial intelligence in local industry. We don’t have it easy. In the 2020 ranking of the 100 most promising AI startups, there are only six within the EU: two German, two Swedish, one French and one Spanish named Sherpa, who makes a voice assistant in real time. The average of the old continent rises a little if we add one in Switzerland and six in the United Kingdom, but it is still very low compared to that of China and the United States.
But the opportunity exists. We know that Chinese and American solutions come at a price in data, and there are incentives to avoid them, especially by administrations, public services and, especially, defense. We could also be pioneers in non-extractive technologies that many countries are eager to buy. However, the Commission’s proposal lacks sufficient incentives to generate such an alternative. It is designed to act only “when fundamental rights are at stake”, leaving ample room for innovation.
The Commission proposes to regulate only those things it considers high risk , which are not song or headline recommendation algorithms or which book to buy next. On the one hand, there are the systems that can cause harm, criminal behavior, damage or discrimination to citizens in a hidden or malicious way; on the other, automatic and remote biometric identification systems, although with a wide range of exceptions.
For example, it prohibits AI programs that allow you to change one person’s face for another in a video , popularly known as Deep Fake. Although they only appear in the media when they are used in political campaigns or to resurrect pharaohs in beer advertisements, 99% of incidents are linked to the harassment and humiliation of ex-girlfriends and classmates or workmates, a genre so common that it has its own own label: revenge porn.
It also prohibits social credit systems, such as the one used by the Chinese government to punish or reward citizens , although it is not clear how it intends to monitor those that are already done in an opaque way, in the case of the algorithms that select candidates to receive a kidney or a job or calculate the price of insurance or a mortgage. They are systems trained on past decisions and therefore prone to inherit economic, racial, and sexual biases and automate silent discrimination. At the moment, the proposal is designed to control the companies that develop the software , which will have to be willing to open the code for inspection since that code is auditable. An exciting first in a world dominated by the black boxes of Google, Facebook, Amazon, Apple, IBM and Microsoft.
More worrisome is the proposal on automatic facial recognition systems in real time. In principle, the police would not be able to carry out recognition sweeps in real time of all the people entering and leaving a stadium, a concert or a demonstration . Unless you are looking for a missing child or a suspect of terrorism, murder or fraud. It is too large a trawl that offers opportunities for abuse in vulnerable populations.
